Humanoid robots need mountains of real-world manipulation data, so the industry has invented a new job: strap a phone to your face and do chores like you’re livestreaming for the robot apocalypse.
According to MIT Technology Review, companies are paying contractors in dozens of countries to record everyday tasks (folding laundry, washing dishes, cooking) to sell as training data to humanoid-robot builders.
Who Wins
Robot makers: They get the only thing more scarce than capital, real manipulation footage with enough variation to (maybe) generalize beyond a single kitchen and one very tired ironing board.
Data brokers: If you can recruit, vet, review, label, and package “hands doing stuff” at scale, you’ve basically become the toll booth on the road to household robots.
Workers (sometimes): MIT Tech Review reports workers earning around $15/hour in Nigeria, which can be meaningful local income even when the work is repetitive, awkward, and socially baffling.
Who Loses
Privacy (everyone’s): Even with faces blurred, a home is a fingerprint. The footage captures interiors, routines, possessions, and bystanders. Consent gets fuzzy fast in shared spaces.
Safety-by-default: A robotics expert at ASTM (quoted by MIT Tech Review) warns that people don’t always do tasks the “safe” way at home, which can bake bad habits into training data unless aggressively filtered.
Quality control: When the business model is “more hours,” review becomes the bottleneck. The same scale that makes the dataset valuable also makes it harder to police.
Who Gets Weird Incentives
The whole stack: If the metric is “hours of footage,” you will get footage. Whether you get useful, safe, consented footage is… a separate KPI nobody wants on the investor slide.
The Droid Brief Take
This is the least glamorous part of humanoid robotics, which is exactly why it’s the most honest. Your robot butler isn’t being trained by genius breakthroughs, it’s being trained by someone in a studio apartment trying to invent a 37th way to fold a T-shirt without showing their kid’s face.
What to Watch
Consent tooling: Do these companies build real mechanisms for deletion, audit, and bystander consent, or do they just keep saying “don’t film personal info” and hope reality behaves?
Data quality standards: If training data becomes a safety input, standards bodies will get dragged into the loop, whether anyone likes it or not.
The hidden human layer: MIT Tech Review frames this as a gig economy. The more important question is how much “autonomy” depends on remote supervision, annotation, and cleanup humans you never see.
Sources
MIT Technology Review — “The gig workers who are training humanoid robots at home”